%0 = "tfl.sub"(%arg0, %cst) {fused_activation_function = "NONE"} : (tensor<*xi32>, tensor<i32>) -> tensor<*xi32>
  %1 = tfl.add %arg1, %arg1 {fused_activation_function = "NONE"} : tensor<*xf32>
  func.return %0, %1 : tensor<*xi32>, tensor<*xf32>
}

func.func @main(%arg0: tensor<i32>, %arg1: tensor<1xf32>) -> tensor<1xf32> {
  %0:2 = "tfl.while"(%arg0, %arg1) ({
  ^bb0(%arg2: tensor<*xi32>, %arg3: tensor<*xf32>):

func.func @replicate_with_packed_input(%arg0: tensor<i1>, %arg1: tensor<i1>) {
  %0:4 = tf_executor.graph {
    %1:5 = tf_executor.island {
      %2:4 = tf_device.replicate(%arg0 as %arg2: tensor<i1>, %arg1 as %arg3: tensor<i1>)
          {n = 2 : i32, _packed_input_indices = [0, 1]} {
        %3 = "tf.opA"(%arg2) : (tensor<i1>) -> tensor<f32>
        %4 = "tf.opB"(%arg3) : (tensor<i1>) -> tensor<i32>

  func.func @inference_Fn_101680(%arg0: tensor<*x!tf_type.string>, %arg1: tensor<*x!tf_type.string>, %arg2: tensor<*xi32>, %arg3: tensor<*xi32>) -> tensor<*xi32> {

  %0 = func.call @f(%arg0, %arg1, %arg0, %arg1) : (f32, f32, f32, f32) -> f32
  return %0 : f32
}

// -----

// CHECK-LABEL: leaves_while_loops_alone
func.func @leaves_while_loops_alone(%arg0: tensor<i32>, %arg1: tensor<f32>) {
  // CHECK: While{{.*}}(%arg0, %arg1)
  %0, %1 = "tf.While"(%arg0, %arg1) {body = @body, cond = @cond, is_stateless = false} :
      (tensor<i32>, tensor<f32>) -> (tensor<i32>, tensor<f32>)

A#loc9 = loc(fused[#loc1, #loc2, #loc3])\0Amodule {\0A  func.func @host_func(%arg0: tensor<3360x?xi32> loc(fused[#loc1, #loc2, #loc3]), %arg1: tensor<3xi32> loc(fused[#loc1, #loc2, #loc3]), %arg2: tensor<i32> loc(fused[#loc1, #loc2, #loc3]), %arg3: tensor<2xi32> loc(fused[#loc1, #loc2, #loc3])) -> (tensor<1x1120x?xi32>, tensor<1x1120x?xi32>, tensor<1120x?xi32>, tensor<2xi32>) {\0A    %0 = \22tf.Reshape\22(%arg0, %arg1) {_xla_outside_compilation = \220\22} : (tensor<3360x?xi32>, tensor<3xi32>) ->...

  func.func @value_used_in_multiple_same_scale_ops(%arg0: tensor<*xf32>, %arg1: tensor<*xf32>, %arg2: tensor<*xf32>, %arg3: tensor<*xf32>) -> (tensor<*xf32>, tensor<*xf32>) {
    %cst = "tf.Const"() {device = "", value = dense<2> : tensor<i32>} : () -> tensor<i32>
    %0 = "quantfork.stats"(%arg0) {layerStats = dense<[-0.83811146, 62.4960899]> : tensor<2xf32>} : (tensor<*xf32>) -> tensor<*xf32>

}

func.func @testTanh(%arg0: tensor<10x10xf32>) -> tensor<10x10xf32> {
  // CHECK: _arithmetic_count = 6400 : i64
  %0 = "tfl.tanh"(%arg0) : (tensor<10x10xf32>) -> tensor<10x10xf32>
  func.return %0 : tensor<10x10xf32>
}

func.func @testAddN(%arg0: tensor<10x10xf32>, %arg1: tensor<10x10xf32>, %arg2: tensor<10x10xf32>) -> tensor<10x10xf32> {
  // CHECK: _arithmetic_count = 200 : i64

// CHECK-LABEL: bias_add
func.func @bias_add(%arg0: tensor<1x10x10x32xf32>, %arg1: tensor<32xf32>) -> tensor<1x10x10x32xf32> {
  %0 = "tf.BiasAdd"(%arg0, %arg1) {T = "tfdtype$DT_FLOAT", data_format = "NHWC"} : (tensor<1x10x10x32xf32>, tensor<32xf32>) -> tensor<1x10x10x32xf32>
  func.return %0 : tensor<1x10x10x32xf32>

}

// -----

func.func @testConv2DCPU(tensor<256x32x32x3xf32>, tensor<16x3x3x3xf32>, tensor<16xf32>) -> tensor<256x32x32x16xf32> {
^bb0(%arg0: tensor<256x32x32x3xf32>, %arg1: tensor<16x3x3x3xf32>, %arg2: tensor<16xf32>):
  // CHECK: tac.cost = 0x4D5C0000

// CHECK-SAME: [[ARG0:%.*]]: tensor<f32> {tf.aliasing_output = 1 : i64},
// CHECK-SAME: [[ARG1:%.*]]: tensor<f32> {tf.aliasing_output = 2 : i64},
// CHECK-NOT: [[ARG2:%.*]]:  tensor<f32> {tf.aliasing_output = 0 : i64}
func.func @device_func_5(%arg0: tensor<f32>, %arg1: tensor<f32>, %arg2: tensor<f32>) -> (tensor<f32>, tensor<f32>, tensor<f32>) {
  func.return %arg2, %arg1, %arg0 : tensor<f32>, tensor<f32>, tensor<f32>